Limits of Predictability for Large-Scale Urban Vehicular Mobility

Abstract

Key challenges in vehicular transportation and communication systems are understanding vehicular mobility and utilizing mobility prediction, which are vital for both solving the congestion problem and helping to build efficient vehicular communication networking. Most of the existing works mainly focus on designing algorithms for mobility prediction and exploring utilization of these algorithms. However, the crucial questions of how much the mobility is predictable and how the mobility predictability can be used to enhance the system performance are still the open and unsolved problems. In this paper, we consider the fundamental problem of the predictability limits of vehicular mobility. By using two large-scale urban city vehicular traces, we propose an intuitive but effective model of areas transition to describe the vehicular mobility among the areas divided by the city intersections. Based on this model, we examine the predictability limits of large-scale urban vehicular networks and obtain the maximal predictability based on the methodology of entropy theory. Our study finds that about 78%-99% of the location and above 70% of the staying time, respectively, are predicable. Our findings thus reveal that there is strong regularity in the daily vehicular mobility, which can be exploited in practical prediction algorithm design.